How ARMA Models Enable Accurate Time Series Forecasting

ARMA Model Forecasting

Step forecasting of a time series estimates future random variables based on past observed values. The estimator is a linear combination of the series' past observations.

Forecasting a Series

The recursive forecasting formula shows that the forecast depends only on a finite number of previous values of the series, which is a characteristic of time‑series prediction.

Forecasting MA(q) Series

For an MA(q) process, the forecast can be derived by defining a forecast vector and establishing the recursive relationship between the series and its past innovations. The recursive initial values can be chosen small because model invertibility guarantees asymptotic stability; as the horizon grows, the impact of initial errors diminishes.

For an ARMA series, knowing the past observations and the model parameters allows recursive computation of future values. The same forecast vector (as in equation 18.25) is used, and the recursive forecast formulas can be proved under the invertibility condition.

In practice, model parameters are unknown and must be estimated. Once a time‑series model is fitted, the estimated parameters replace the theoretical ones, and the recursive method described above is applied to generate forecasts.